Previous primate electrophysiological and human psychophysical studies have provided great insights into the mechanisms of color vision. However, few hypotheses have been validated directly because of the complexity and variability of vertebrate neural circuits as well as the technical difficulty when attempting to establish causality. To circumvent these problems, we use the Drosophila visual system as a model to study color vision. True color vision and high-order color vision functions have been demonstrated in many insects including bees and flies. We use a combination of molecular, genetic, histological, and behavioral approaches to determine the synaptic circuits involved in color vision and to identify the critical neurons that process color information. To determine the synaptic circuits of color vision, we combined molecular genetic and histological approaches. We are currently focusing on the medulla neuropil, which is analogous to the inner plexiform layer of vertebrate retina. The medulla is innervated by the chromatic photoreceptors R7 and R8 as well the first-order interneurons (L1-5) of the achromatic photoreceptor R1-R6. Approximately sixty different types of medulla neurons process the visual information carried by these afferents. To overcome the cellular complexity, we devised a divide and conquer strategy and subdivided the medulla neurons into several subclasses based on their use of neurotransmitters and receptors. We took the advantage of the finding that fly photoreceptor neurons utilize histamine as neurotransmitter. Therefore the first-order interneurons must express the histamine receptor (histamine-gated chloride channel or HisCl) in order to respond to histamine signal. Based on HisCl expression expression, we identified an amacrine neuronal type, Dm8, which relays R7 signal to projection neurons and three types of projection neurons, Tm5, Tm9 and Tm20, that relay photoreceptor signal to the higher visual center, the lobula. Using the promoters of various transcription factors, neurontransmitter transporters and synthesis enzymes, we further divided these neurons into subgroups. We found that the Tm5 can be further divided into three subtypes, Tm5a, Tm5b, and Tm5c, each of which has a unique axonal projection and dendritic arborization pattern correlated with its distinct gene expression profile. To characterize the dendritic branching pattern of different projection neurons, we developed an imaging technique called dual-view imaging, which generates high-resolution 3D images by combining two confocal image stacks collected in orthogonal orientations. Unlike typical confocal images which have low axial resolution, the dual view images are isotropic. Using this technique, we characterized the dendritic branching pattern of the Tm9 neurons as well as the Tm2 neurons, the third order interneurons involved in motion detection. We found that while different Tm neurons have stereotyped dendritic branching patterns, the detailed branching topology varies greatly from one neuron to another within a single neuronal type. We are now combining this technique with serial EM reconstruction to determine whether the connectivity is type invariant. To determine the functions of different types of medulla neurons, we used genetic technique to inactivate or restore their synaptic functions and examined the behavioral consequences. This procedure allows us to assign specific functions to different neuronal subtypes, therefore establishing causality. The specific types of neurons amenable to this approach are limited by the specificity and diversity of genetic drivers available. The split-Gal4 system developed recently combines two promoters to enhance specificity but the number of available drivers is rather limited. To overcome this problem, we developed a concatenated expression system, which is based on our split-LexA system and is compatible with existing Gal4 drivers abundantly available. Using this new expression system as well as the original split-Gal4 system, we have generated many genetic drivers that targeted specific medulla neurons. With these cell-specific drivers, we have begun to examine systematically whether a specific neuron subtype is required or sufficient for color and motion detection. We have found that the amacrine neuron Dm8 is specifically required and sufficient for animals preference to UV light over green light but not for motion detection. Conversely, the lamina neurons L1 and L2 are only required for motion detection but not for color vision. In summary, our study validates the validity of our approach and reveals that different neuron subtypes subserve distinct visual functions.